abstract |
The invention discloses an online multi-target tracking method based on multi-function aggregation and tracking simulation training, and belongs to the technical field of computer vision. The method designs a network structure that aggregates multiple functional modules, and uses the fusion mode of ECC and Kalman filter as the motion model. At the same time, the application adopts an end-to-end training method, which uses the predicted position of the target and the actual position to expand. Training data, and finally, the appearance extraction module in the network is trained with an indicator loss that incorporates the historical appearance information of the target. The method of the invention can improve the compatibility between various functional modules and the adaptability to the multi-target tracking task, improve the identity preservation ability of the model, and has better online multi-target tracking performance. |